Re: [PATCH 4/5] break up slow_virt_to_phys()

[Dave Hansen <dave@sr71.net>]

On 04/11/2013 05:29 AM, Borislav Petkov wrote:
> On Wed, Apr 10, 2013 at 04:32:54PM -0700, Dave Hansen wrote:
>> +phys_addr_t slow_virt_to_phys(void *virt_addr)
>> +{
>> +	phys_addr_t result;
>> +	int ret;
>> +
>> +	ret = kernel_lookup_vaddr(virt_addr, &result);
>> +	BUG_ON(ret);
> 
> Isn't that BUG_ON still too harsh though? How about WARN_ON instead?
> It would still create a lot of noise so that it gets fixed without
> bringing down the system.

It's harsh for the cases where __pa()'s result never gets used directly,
like for checking against 'high_memory'.  If it gets used (like in
/dev/mem's case) the kernel really is doing something it does not intend
to do.  It's essentially reading or writing garbage.

On the other hand, I guess if we've been letting it do that bad thing
for years, we might as well continue.  We're not causing any _more_ damage.

I'm leaning toward making it a WARN_ON() at the moment. :)